What are the difference between the output['predicts_j3ds'] and the joints got from the SMPL model using output['betas'] and output['thetas'] ?

Open sycamore27 opened this issue 3 years ago • 1 comments

What are the difference between the output['predicts_j3ds'] and the joints got from the SMPL model using output['betas'] and output['thetas'] ? I found the results are different. And I use the joints got from the SMPL model using output['betas'] and output['thetas'], to reproject to the 2D image(the green points) using the estimated K, which is wrong comparing the outputs['pj2d_org'](the red points). test1

Could you please answer my question? Thanks!

Nov 03 '22 17:11 sycamore27

Hi, @sycamore27 , Sorry for the late reply. The output['predicts_j3ds'] is obtained via https://github.com/Arthur151/ROMP/blob/master/simple_romp/romp/smpl.py using natural SMPL model with extra joints defined at https://github.com/Arthur151/ROMP/blob/master/romp/lib/constants.py

Please note that we apply a root alignment operation at https://github.com/Arthur151/ROMP/blob/f3c76725fff3f9cd0d3a90721315dcd5f23c3f58/simple_romp/romp/smpl.py#L102

Nov 23 '22 02:11 Arthur151